copilot: Indicate whether a request is initiated by an agent to Copilot API (#33895)

Per [GitHub's documentation for VSCode's agent
mode](https://docs.github.com/en/copilot/how-tos/chat/asking-github-copilot-questions-in-your-ide#agent-mode),
a premium request is charged per user-submitted prompt. rather than per
individual request the agent makes to an LLM. This PR matches Zed's
functionality to VSCode's, accurately indicating to GitHub's API whether
a given request is initiated by the user or by an agent, allowing a user
to be metered only for prompts they send.

See also: #31068

Release Notes:

- Improve Copilot premium request tracking

crates/copilot/src/copilot_chat.rs

@@ -528,6 +528,7 @@ impl CopilotChat {
 
     pub async fn stream_completion(
         request: Request,
+        is_user_initiated: bool,
         mut cx: AsyncApp,
     ) -> Result<BoxStream<'static, Result<ResponseEvent>>> {
         let this = cx
@@ -562,7 +563,14 @@ impl CopilotChat {
         };
 
         let api_url = configuration.api_url_from_endpoint(&token.api_endpoint);
-        stream_completion(client.clone(), token.api_key, api_url.into(), request).await
+        stream_completion(
+            client.clone(),
+            token.api_key,
+            api_url.into(),
+            request,
+            is_user_initiated,
+        )
+        .await
     }
 
     pub fn set_configuration(
@@ -697,6 +705,7 @@ async fn stream_completion(
     api_key: String,
     completion_url: Arc<str>,
     request: Request,
+    is_user_initiated: bool,
 ) -> Result<BoxStream<'static, Result<ResponseEvent>>> {
     let is_vision_request = request.messages.iter().any(|message| match message {
       ChatMessage::User { content }
@@ -707,6 +716,8 @@ async fn stream_completion(
       _ => false,
   });
 
+    let request_initiator = if is_user_initiated { "user" } else { "agent" };
+
     let mut request_builder = HttpRequest::builder()
         .method(Method::POST)
         .uri(completion_url.as_ref())
@@ -719,7 +730,8 @@ async fn stream_completion(
         )
         .header("Authorization", format!("Bearer {}", api_key))
         .header("Content-Type", "application/json")
-        .header("Copilot-Integration-Id", "vscode-chat");
+        .header("Copilot-Integration-Id", "vscode-chat")
+        .header("X-Initiator", request_initiator);
 
     if is_vision_request {
         request_builder =

crates/language_models/src/provider/copilot_chat.rs

@@ -30,6 +30,7 @@ use settings::SettingsStore;
 use std::time::Duration;
 use ui::prelude::*;
 use util::debug_panic;
+use zed_llm_client::CompletionIntent;
 
 use super::anthropic::count_anthropic_tokens;
 use super::google::count_google_tokens;
@@ -268,6 +269,19 @@ impl LanguageModel for CopilotChatLanguageModel {
             LanguageModelCompletionError,
         >,
     > {
+        let is_user_initiated = request.intent.is_none_or(|intent| match intent {
+            CompletionIntent::UserPrompt
+            | CompletionIntent::ThreadContextSummarization
+            | CompletionIntent::InlineAssist
+            | CompletionIntent::TerminalInlineAssist
+            | CompletionIntent::GenerateGitCommitMessage => true,
+
+            CompletionIntent::ToolResults
+            | CompletionIntent::ThreadSummarization
+            | CompletionIntent::CreateFile
+            | CompletionIntent::EditFile => false,
+        });
+
         let copilot_request = match into_copilot_chat(&self.model, request) {
             Ok(request) => request,
             Err(err) => return futures::future::ready(Err(err.into())).boxed(),
@@ -276,7 +290,8 @@ impl LanguageModel for CopilotChatLanguageModel {
 
         let request_limiter = self.request_limiter.clone();
         let future = cx.spawn(async move |cx| {
-            let request = CopilotChat::stream_completion(copilot_request, cx.clone());
+            let request =
+                CopilotChat::stream_completion(copilot_request, is_user_initiated, cx.clone());
             request_limiter
                 .stream(async move {
                     let response = request.await?;